The future of machine translation and how to use it properly

Edit on 22nd July 2021: Autofill is now called Prefill.

There are a lot of myths about MT(machine translation). Many hyping titles like "The MT will replace people" appeared when AI algorithms were added to the translating systems. One such famous example is ChatGPT which we described in a separate article in detail. This blog post aims to break the myths and show how businesses and translators can benefit from Machine Translation systems to accelerate translation.

What is machine translation?

Machine Translation (MT) is when artificial intelligence or other algorithms translate text without human involvement. Generally, there are three types of Machine Translation to distinguish:

1. Rule-based machine translation (RBMT)

• Data - relies on linguistic rules and bilingual dictionaries for every language pair.
• Translation - uses rule sets to transfer the grammatical structure of source content into the target language.

2. Statistical machine translation (SMT)

• Data - instead of extensive databases, it uses statistical translation models from analyzing training data.
• Translation - is selected from training data with algorithms to choose the most commonly appearing words.

3. Neural machine translation (NMT)

• Data - it uses Ai algorithms for deep learning on the prepared texts.
• Translation - finds dependencies between a sequence of words based on the learned information to provide the best-fit result.

How to use Lingohub machine translation the right way

Misusing machine translation might harm your business because grammatically incorrect and faulty translations will likely lead to a bad user experience. In these 4 use cases, we described how your business and translators could profit from machine translation usage:

#1 Use machine translation if no translation memory suggestions are available

At Lingohub, we provide a powerful translation memory tool that suggests translations based on previous efforts or from imported files. During localization, the translation memory provides recommendations if it finds fuzzy or exact matching between new text and the content in the memory. If there are no suggestions for a text, use the MT as a perfect tool for fast pre-translating.

#2 Machine translation for workflow optimization

Situations, when the project requires fast-translated text can be faced often. An example, when you need a drafted translation to check how the design will work with the new language, waiting for the fully ready content is time-wasting. Lingohub provides a super-convenient chain of actions to make this process faster than you can imagine.

• Step 1 - use a Figma plugin to transfer the source text from the design template to the Lingohub Organization.
• Step 2 - use the prefill tool; all your source text will be translated automatically.
• Step 3 - push the text translated via machine translation back to Figma and evaluate the required design changes. The prefill function can be used for all translations to save the translators time and reduce costs. Your employees will never start from scratch and only do the post-edit translations.

#3 Machine translation for user-generated content

Frequent changes and updates are characteristics of user-generated content. Translating all such texts by professional translators will be expensive, while MT offer is a less costly and quite reliable alternative:

• 5-20% of machine translation suggestions can be final translations.
• Roughly 40% of suggestions can be published after post-editing.
• MT provides data to autocomplete up to 80% of texts.

Machine translation accuracy statistic

One of the most popular MT engines today is Google Translate, with over 1 billion users, so let's look at how successfully it can cope with different language pairs. In 2021 UCLA Medical Center researched this question. They have used instructions for patients as test text and translated them into Spanish, Chinese, Tagalog, Korean, Armenian, and Farsi. The results were the following:

• Spanish (94%),
• Tagalog (90%),
• Korean (82.5%),
• Chinese (81.7%),
• Farsi (67.5%),
• Armenian (55%). In conclusion, some language pairs are more popular, and the NMT can better understand the context, while others have much lower success. The professional translator, in the best case, the native speaker, is an essential part of successful content translation.

The future of machine translation

With AI and neural network development, machine translation has become more competent and smarter. We can see this in the ChatGPT and GT examples that work on the Transformer algorithms base. The more information they get - the better they understand the context and provide relevant results. Nowadays, it can speed up the translation 3-4 times, and maybe one day, artificial intelligence will cover all the popular languages in the same high quality. Want to learn more about how MT started and developed? Highly recommended to read this article.

Summing it up

Machine translation does not yet keep up with human translators. However, NMT is an essential step in the right direction to enhance its capabilities. Using MT is the right way will accelerate your translation speed, minimize per-word costs and relieve your translators. Try it for yourself and start your free Lingohub trial today!